context-packager

name	context-packager
description	Package and compress the context needed for an AI-assisted analysis session. Use when starting a complex analysis with Claude to minimise token usage, reduce repetition, and establish shared understanding upfront.

When to use

• Starting a multi-turn AI analysis session where context will be referenced repeatedly
• Restarting a conversation and need to rebuild context efficiently
• Sharing context with a collaborator who will continue an AI-assisted analysis
• When an AI session keeps losing context and producing inconsistent results

Process

1. Identify context layers — use references/context_layering_guide.md to determine what layers are needed: business context, data schema, metric definitions, analytical constraints, and output format preferences
2. Consolidate sources — use scripts/context_bundler.py to merge schema docs, metric definitions, and business rules into a single structured file
3. Verify token budget — estimate token count; trim to fit within budget while preserving the highest-signal context
4. Assess quality — score the package against references/context_quality_rubric.md; ensure it covers the minimum viable context for the analysis
5. Craft the task statement — write a clear opening prompt that references the packaged context and states the analytical goal
6. Save for reuse — store the package for future sessions on the same dataset or analysis area

Inputs the skill needs

• Required: description of the analytical goal
• Required: source materials (schema docs, metric definitions, reports, data samples)
• Required: token budget ceiling

Output

• Unified context file (structured markdown)
• Token estimate
• Context quality score
• Formatted prompt ready for use with Claude